MLGraph is a great abstract definition of an ML pipeline independent of runtime.
It'd be beneficial to define an MLGraph once, with the ability to run it in batch (Kubeflow Pipelines), streaming or serving (Seldon, KFServing, etc). Data may be bound and accessed between components through a data abstraction layer (DAL), see https://github.com/zillow/intake-dal.
- Has there been thought given for MLGraph to run on KubeFlow Pipelines?
- Are there plans or thoughts to have a Pythonic SDK to author an MLGraph?
Thank you, I look forward to discussing and working with you on this!
MLGraph is a great abstract definition of an ML pipeline independent of runtime.
It'd be beneficial to define an MLGraph once, with the ability to run it in batch (Kubeflow Pipelines), streaming or serving (Seldon, KFServing, etc). Data may be bound and accessed between components through a data abstraction layer (DAL), see https://github.com/zillow/intake-dal.
Thank you, I look forward to discussing and working with you on this!